Glossary
Terms in bold are written as Studio shows them, in Spanish, followed by their English translation in parentheses. Where the interface also uses another name for a term, the parentheses say so (for example, "corrida").
| Term | Meaning |
|---|---|
| Agente (Agent) | Assistant configured with a model, instructions, parameters, tools and knowledge. It can be individual or a team. |
| Ajustes avanzados (Advanced settings) | An agent's creativity, maximum response length, message history and tool calls. |
| Canal (Channel) | Connection through which information enters or leaves a workflow: webhook, WhatsApp, internal input and output. |
| Catálogo de modelos (Model catalog) | List of the models available on the Client's infrastructure and their storage. |
| Cliente (Client) | Organization you belong to. It contains members, agents, workflows, labs and credentials, isolated from other Clients. |
| Credencial (Credential) | Key for using an external provider or InferNode. Agents and workflows reference it without exposing the secret. |
| Editor (lab) | Role of whoever receives a shared lab: they can open it and work in it, but not share or delete it. |
| Ejecución (Run; in the interface, also "corrida") | One pass of a workflow, with its status, source, trace and cost. |
| Reasoning effort | How much the model reasons before responding: Instantáneo (Instant), Bajo (Low), Medio (Medium) or Alto (High). You choose it in the test chat. |
| Frida | Saptiva AI control plane that provisions and manages the labs. |
| Fuente (Source) | Document or connector that an agent queries through retrieval (RAG). Private by default to whoever uploads it. |
| Herramienta (Tool) | Capability that an agent can invoke in a response. The ones the selector offers depend on the installation; today they are Web Search and Leer CSV y Excel (Read CSV and Excel). |
| InferNode | Model gateway: an OpenAI-compatible API for all models, with limits, guardrails and metering per Client. |
| Laboratorio (Lab) | Isolated environment with CPU, memory, GPU and disk for notebooks, pipelines, training, experiments and serving. |
| Miembro (Member) | Person with an account in a Client. |
| Diccionario de Campos (Field Dictionary) | AI node that extracts structured fields from a document with templates per type (invoice, receipt, articles of incorporation) and regex before the model. |
| Dictaminación (Ruling) | AI node that evaluates a case file against natural-language rules and consistency across documents, and delivers a ruling. |
| Nodo (Node) | Step of a workflow. An agent is a type of node. |
| Human Review | Node that pauses the run until a person approves or rejects the fields; it does not branch. |
| Lab panel | Interface that opens with Abrir (Open) from the list of labs. It is in English, and the docs keep its names as they are. |
| Notebooks | Section of the lab panel: Jupyter, VS Code or RStudio environments. |
| Volumes | Section of the lab panel: the lab's disks. The Workspace Volume is the one that persists your work. |
| Pipelines | Section of the lab panel: multi-step processes that you launch and repeat. |
| Runs | Section of the lab panel: each run of a pipeline. |
| Experiments | Section of the lab panel: grouping of runs to compare results. |
| Endpoints | Section of the lab panel: models published to receive requests. |
| Plantilla (Template) | Prebuilt workflow for a use case, a starting point for your own. |
| Rol (Role) | Permission level within the Client. Studio writes them in English and lowercase: owner, admin, member, viewer. When inviting, you can only choose member and admin. |
owner | Role of whoever created the Client. It behaves like admin and is the only one that can transfer or delete the Client. SSO never changes it. |
viewer | Read-only role. It does not appear in the invitation selector. |
| Super Admin | Platform role, for Saptiva staff. It manages Clients and, when it enters one, sees the same as an administrator of that Client. It is not granted from Accesos (Access). |
| Redimensionar (Resize) | Change the size or resources of an existing lab. It applies live if what is running fits in the target; if not, it asks for consent to stop notebooks. The disk only goes up. |
| Request / limit | What a container needs guaranteed to start and the most it can consume. The lab budget is counted against the limits. |
| Talla (Size) | Predefined size of a lab across five axes (CPU, memory, cards, VRAM per card, disk). It works as a budget shared by everything that runs inside. A lab whose disk ended up above the plan is no longer a size and shows its resources as numbers. |
| Tarjetas (Cards) | How many physical GPU cards a lab notebook can use at once. Together with the VRAM per card they form the GPU budget: N cards with M of VRAM each. |
| Traza (Trace) | List of events of a run: nodes, tool calls, model invocations and errors. |
| Trial | Each attempt of a hyperparameter experiment, with its parameters and metrics. |
| VRAM | Memory of the GPU card. The platform splits it into fractions among labs; each process sees only the amount of its size and, in a lab with several cards, the amount of each card separately. |
| Versión (Version) | Published snapshot of a workflow. The draft is edited separately and does not affect the version in production. |
| Webhook | Public URL of a published workflow that triggers it over HTTP, protected with a token. |
| Flujo (Workflow) | Sequence of nodes that processes inputs and produces outputs; it is published and triggered by webhook, WhatsApp or API. Studio calls it Flujos (Workflows) in the menu; the agent card and the API paths call it workflows. |